Hardware Software Codesign of Embedded System CPSC489-501 Rabi Mahapatra Mahapatra - Texas A&M - Fall 00 1 Today s topics Course Organization Introduction to HS-CODES Codesign Motivation Some Issues on Codesign of Embedded System Mahapatra - Texas A&M - Fall 00 2 1
Course Organization Lectures: HRBB 126, MWF 11:30-12:20 Laboratory: HRBB 218, TBD (Some Two hours except MTWR afternoons) Teaching Assistant: Brian G. Murray bgm0975@cs.tamu.edu Send emails for alias Instructor s office Hours: By appointment Contact: rabi@cs.tamu.edu, 5-5787 Mahapatra - Texas A&M - Fall 00 3 Course organization: Gradings Two tests: 50% Labs: 20% Projects: 20% Assignments/Term papers: 10% Projects and Labs: Team work Term papers: Individual responsibility Mahapatra - Texas A&M - Fall 00 4 2
Course policies Required to access the course web page for relevant info during the semester class attendance is required use emails for effective communication with TA and Instructor all assigned papers are required materials follow lab and univ rules Mahapatra - Texas A&M - Fall 00 5 Topics to be covered (order and details may change) 1. Codesign overview 2. Models and methodologies of system design 3. Hardware software partitioning and scheduling 4. Cosimulation, synthesis and verifications 5. Architecture, Interface and reconfiguration 6. System on chip 7. Application specific processors (DSP) 8. Codesign tools and case studies Mahapatra - Texas A&M - Fall 00 6 3
Texts Staunstrup and Wolf Ed. Hardware Software codesign: principles and practice, Kluwer Publication, 1997 Gajski, Vahid, Narayan and Gong, Specification, Design of Embedded Systems, Prentice Hall, 1994 Suggested papers in the class web Mahapatra - Texas A&M - Fall 00 7 Reading assignment Giovanni De Micheli and Rajesh Gupta, Hardware/Software co-design, IEEE Proceddings, vol. 85, no.3, March 1997, pp. 349-365. Mahapatra - Texas A&M - Fall 00 8 4
Introduction Digital systems designs consists of hardware components and software programs that execute on the hardware platforms Hardware-Software Codesign? Mahapatra - Texas A&M - Fall 00 9 Microelectronics trends Better device technology reduced in device sizes more on chip devices > higher density higher performances Mahapatra - Texas A&M - Fall 00 10 5
Microelectronics trends Higher degree of integration increased device reliability inclusion of complex designs Mahapatra - Texas A&M - Fall 00 11 Digital Systems Judged by its objectives in application domain Performance Design and Manufacturing cost Ease of Programmability Mahapatra - Texas A&M - Fall 00 12 6
Digital Systems Judged by its objectives in application domain Performance Design and Manufacturing cost Ease of Programmability It depends on both the hardware and software components Mahapatra - Texas A&M - Fall 00 13 Hardware/Software Codesign A definition: Meeting System level objectives by exploiting the synergism of hardware and software through their concurrent design Mahapatra - Texas A&M - Fall 00 14 7
Concurrent design Traditional design flow start Concurrent (codesign) flow start HW SW Designed by independent groups of experts HW SW Designed by Same group of experts with cooperation Mahapatra - Texas A&M - Fall 00 15 Codesign motivation Trend toward smaller mask-level geometries leads to: Higher integration and cost of fabrication. Amortize hardware design over large volume productions Suggestion: Use software as a means of differentiating products based on the same hardware platform. Mahapatra - Texas A&M - Fall 00 16 8
Story of IP cores What are these IP Cores? Predesigned, preverified silcon circuit block, usually containing 5000 gates, that can be used in building larger application on a semiconductor chip. Mahapatra - Texas A&M - Fall 00 17 Story of IP cores Complex macrocells implementing instruction set processors (ISP) are available as cores Hardware (core) Software (microkernels) Are viewed as intelectual property Mahapatra - Texas A&M - Fall 00 18 9
IP core reuse Cores are standardized for reuse as system building blocks Rationale: leveraging the existing software layers including OS and applications in ES Results: 1. Customized VLSI chip with better area/ performance/ power trade-offs 2. Systems on Silicon Mahapatra - Texas A&M - Fall 00 19 Hardware Programmability Traditionally Hardware used to be configured at the time of manufacturing Software is variant at run time The Field Programmable Gate Arrays (FPGA) has blurred this distinction. Mahapatra - Texas A&M - Fall 00 20 10
FPGAs FPGA circuits can be configured on-the-fly to implement a specific software function with better performance than on microprocessor. Mahapatra - Texas A&M - Fall 00 21 FPGAs FPGA circuits can be configured on-the-fly to implement a specific software function with better performance than on microprocessor. FPGA can be reprogrammed to perform another specific function without changing the underlying hardware. Mahapatra - Texas A&M - Fall 00 22 11
FPGAs FPGA circuits can be configured on-the-fly to implement a specific software function with better performance than on microprocessor. FPGA can be reprogrammed to perform another specific function without changing the underlying hardware. This flexibility opens new applications of digital circuits. Mahapatra - Texas A&M - Fall 00 23 Reduce time to market Why codesign? Mahapatra - Texas A&M - Fall 00 24 12
Why codesign? Reduce time to market Achieve better design Explore alternative designs Good design can be found by balancing the HW/SW Mahapatra - Texas A&M - Fall 00 25 Why codesign? Reduce time to market Achieve better design Explore alternative designs Good design can be found by balancing the HW/SW To meet strict design constraint power, size, timing, and performance trade-offs safety and reliability system on chip Mahapatra - Texas A&M - Fall 00 26 13
Distinguishing features of digital system Interrelated criteria for a system design Hardware Technology Application Domain Level of Integration Degree of Programmability Mahapatra - Texas A&M - Fall 00 27 Application Domains General purpose computing system usually self contained and with peripherals Information processing systems Mahapatra - Texas A&M - Fall 00 28 14
Application Domains General purpose computing system usually self contained and with peripherals Information processing systems Dedicated control system part of the whole system, Ex: digital controller in a manufacturing plant also, known as embedded systems Mahapatra - Texas A&M - Fall 00 29 Embedded System Uses a computer to perform certain functions Conceived with specific application in mind examples: dash controller in autombiles, remote controller for robots, answering machines, etc. Mahapatra - Texas A&M - Fall 00 30 15
Embedded Systems Uses a computer to perform certain functions Conceived with specific application in mind examples: dash controller in autombiles, remote controller for robots, answering machines, etc. User has limited access to system programming system is provided with system software during manufacturing not used as a computer Mahapatra - Texas A&M - Fall 00 31 Degree of Programmability Most digital systems are programmed by some software programs for functionality. Two important issues related to programming: who has the access to programming? Level at which programming is performed. Mahapatra - Texas A&M - Fall 00 32 16
Degree of Programmability: Accessibility Understand the role of: End users, application developers, system integrator and component manufacturers. Mahapatra - Texas A&M - Fall 00 33 Degree of Programmability: Accessibility Understand the role of: End users, application developers, system integrator and component manufacturers. Application Developer: System to be retargetable. Mahapatra - Texas A&M - Fall 00 34 17
Degree of Programmability: Accessibility Understand the role of: End users, application developers, system integrator and component manufacturers. Application Developer: System to be retargetable. System Integrator: Ensure compatibility of system components Mahapatra - Texas A&M - Fall 00 35 Degree of Programmability: Accessibility Understand the role of: End users, application developers, system integrator and component manufacturers. Application Developer: System to be retargetable. System Integrator: Ensure compatibility of system components Component Manufactures: Concerned with maximizing product reuse Mahapatra - Texas A&M - Fall 00 36 18
Degree of Programmability Example 1: Personal computer End User: Limited to application level Application Dev.: Language tools, Operating System, highlevel programming environment (off the self components) Component Manf.: Drive by bus standards, protocols etc. Observe that: coalescing the system components due to higher chip density Result: Few but more versatile system hardware components Mahapatra - Texas A&M - Fall 00 37 Example 2. Embedded Systems End user: Limited access to programming Most software is already provided by system integrator who could be application developer too! Mahapatra - Texas A&M - Fall 00 38 19
Level of Programmability Systems can be programmed at application, instruction and hardware levels Application Level: Allows users to specify option of functionality using special language. Example: Programming VCR or automated steering control of a ship Mahapatra - Texas A&M - Fall 00 39 Level of Programmability Instruction-level programmability Most common ways with ISA processors or DSP compilers are used in case of computers In case of embedded systems, ISA is NOT visible Mahapatra - Texas A&M - Fall 00 40 20
Level of programmability Hardware level programmability configuring the hardware (after manufacturing) in the desired way. Example: Microprogramming (determine the behavior of control unit by microprogram) Emulating another architecture by alternation of µp Some DSP implementations too Never in RISC or ISA processors Mahapatra - Texas A&M - Fall 00 41 Programmability Microprogramming Vrs. Reconfigurability Microprogram allows to reconfigure the control unit whereas Reconfigurable system can modify both datapath and controller. Mahapatra - Texas A&M - Fall 00 42 21
Programmability Microprogramming Vrs. Reconfigurability Microprogram allows reconfigure the control unit versus Reconfigurable system can modify both datapath and controller. Reconfigurability increases usability but not the performance of a system. Mahapatra - Texas A&M - Fall 00 43 Performance and Programmability General computing applications: use of superscalar RISC architecture to improve the performance (instruction level programming) Mahapatra - Texas A&M - Fall 00 44 22
Performance and Programmability General computing applications: use of superscalar RISC architecture to improve the performance (instruction level programming) Dedicated Applications: Use of application specific designs (ASICs) for power and performance Neither reusable nor cheap! Mahapatra - Texas A&M - Fall 00 45 Performance and Programmability General computing applications: use of superscalar RISC architecture to improve the performance (instruction level programming) Dedicated Applications: Use of application specific designs (ASICs) for power and performance Neither reusable nor cheap! What if ASICs with embedded cores? Mahapatra - Texas A&M - Fall 00 46 23
Performance and Programmability Any other solutions? How about replacing the standard processors by application specific processors that can be programmed at instruction level (ASIPs). Better power-performance than standard processor? Worse than ASICs Mahapatra - Texas A&M - Fall 00 47 Programmability and Cost:ASIPs Cost can typically be amortized over larger volume than on ASICs (with multiple applications using ASIPs). Ease to update the products and engineering changes through programming the HW, However, includes compiler as additional cost Mahapatra - Texas A&M - Fall 00 48 24
Hardware Technology Choice of hardware to implement the design affects the performance and cost VLSI technology (CMOS or bipolar, scale of integration and feature size etc.) can affect the performance and cost. Mahapatra - Texas A&M - Fall 00 49 Hardware Technology: FPGAs Performance is an order of magnitude less than corresponding non-programmable technology with comparable mask size For high volume production, these are more expensive than ASICs Mahapatra - Texas A&M - Fall 00 50 25
Level of Integration Integration leads to reducing number of parts, which means, increased reliability, reduced power and higher performances But it increases the chip size (cost) and makes debugging more challenging. Standard components for SoC are cores, memory, sensors and actuators. Mahapatra - Texas A&M - Fall 00 51 Embedded System Design Objective Embedded systems: control systems: reactive, real-time function & size: micro controller to high throughput data-processor requires leveraging the components and cores of microprocessors reliability, availability and safety are vital use of formal verification to check the correctness may use redundancy Mahapatra - Texas A&M - Fall 00 52 26
Codesign of ISA ISA is fundamental to digital system design An instruction set permits concurrent design of HW and compiler developments Good ISA design is critical in achieving system usability across applications Goal of codesign in ISP development is to optimize HW utilization by application & OS Mahapatra - Texas A&M - Fall 00 53 Codesign of ISA For high performance in ES, selection of instruction set that matches the application is very important replace the standard core by ASIP ASIPs are more flexible than ASICs but less than ISP ASIP performs better than ISP Mahapatra - Texas A&M - Fall 00 54 27
Challenges with ASIP Compatibility requirement is less important Goal: support specific instruction mixes Price of the flexibility in choosing mixed instruction set is to develop the application specific compiler. CAD of compiler is partly solved problem Mahapatra - Texas A&M - Fall 00 55 Typical codesign process System Description Modeling HW/SW Partitioning Unified representation Software synthesis Interface synthesis Hardware synthesis System integration Instruction set level HW/SW evaluation Mahapatra - Texas A&M - Fall 00 56 28
Steps in Codesign HW-SW system involves specification modeling design space exploration and partitioning synthesis and optimization validation implementation Mahapatra - Texas A&M - Fall 00 57 Steps in codesign Specification List the functions of a system that describe the behavior of an abstraction clearly with out ambiguity. Modeling: Process of conceptualizing and refining the specifications, and producing a hardware and software model. Mahapatra - Texas A&M - Fall 00 58 29
Modeling style Homogeneous: a modeling language or a graphical formalism for presentation partitioning problem used by the designer Heterogeneous: multiple presentations partitioning is specified by the models Mahapatra - Texas A&M - Fall 00 59 Steps in codesign Validation: Process of achieving a reasonable level of confidence that the system will work as designed. Takes different flavors per application domain: cosimulation for performance and correctness Mahapatra - Texas A&M - Fall 00 60 30
Steps in codesign Implementation: Physical realization of the hardware (through synthesis) and of executable software (through compilation). Mahapatra - Texas A&M - Fall 00 61 Partitioning and Scheduling (where and when) A hardware/software partitioning represents a physical partition of system functionality into application-specific hardware and software. Scheduling is to assign an execution start time to each task in a set, where tasks are linked by some relations. Mahapatra - Texas A&M - Fall 00 62 31
Summary: Research areas in codesign Languages Architectural exploration tools Algorithms for partitioning Scheduling SW, HW and interface Synthesis Verification and Testing Mahapatra - Texas A&M - Fall 00 63 32